A stroke high-risk population positioning and evaluation system based on multi-modal data

Abstract

The application discloses a stroke high-risk population positioning and evaluation system based on multi-modal data, relates to the technical field of medical information science, and comprises a stroke prevention and control management platform, which is in communication connection with the following modules, wherein: a multi-modal data acquisition and integration module is used for collecting multi-modal data related to stroke from multiple channels and performing a preprocessing operation on the collected multi-modal data. Through the integration of clinical data, image data, omics data and end health data, the application can comprehensively capture multi-dimensional information related to stroke, especially by simulating hemodynamic characteristics by using a cerebral vascular digital twin and combining a graph neural network to fuse multiple data sources, the application can more accurately identify high-risk populations, significantly improve the accuracy and reliability of risk prediction, and overcome the missed detection problem caused by the dependence on a single data source in traditional methods, thereby providing stronger support for early intervention.

Description

Technical Field

[0001] This invention relates to the field of medical information science and technology, specifically to a stroke high-risk population location and assessment system based on multimodal data. Background Technology

[0002] Stroke, also known as cerebrovascular accident, is a disease caused by problems with blood circulation in the brain, leading to hypoxia and ischemia in brain tissue, which damages brain cells. Stroke is divided into two main categories: ischemic stroke (cerebral infarction) and hemorrhagic stroke (cerebral hemorrhage). It is one of the diseases with high mortality and disability rates worldwide. Studies have shown that high-risk factors for stroke are closely related to lifestyle, environmental factors, genetic factors, and chronic diseases (such as hypertension and diabetes). Therefore, it is particularly important to identify high-risk groups as early as possible and intervene.

[0003] In existing technologies, traditional screening systems rely on electronic health records (EHRs) or questionnaire data, ignoring key information such as imaging and genomics, and cannot reflect fluctuations in the patient's physiological state in real time, resulting in a high rate of missed detection of potentially high-risk groups. Therefore, how to develop a multimodal data fusion framework to integrate clinical, imaging, omics data and digital twin simulation results to improve the accuracy of high-risk group location assessment is the problem to be solved by this invention. To this end, a stroke high-risk group location assessment system based on multimodal data is proposed. Summary of the Invention

[0004] The purpose of this invention is to provide a stroke high-risk population localization and assessment system based on multimodal data, so as to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0006] A stroke high-risk population localization and assessment system based on multimodal data includes a stroke prevention and control management platform, which is communicatively connected to the following modules, wherein:

[0007] The multimodal data acquisition and integration module is used to collect multimodal data related to stroke from multiple channels, including clinical data, imaging data, omics data and end-point health data. It performs preprocessing operations such as cleaning, denoising and standardization on the collected multimodal data, and integrates them into a stroke basic dataset in a unified format to improve data quality and eliminate differences and redundancy between data.

[0008] The digital twin simulation module is used to construct a digital twin of the cerebral blood vessels of stroke patients, simulate the structure and hemodynamic characteristics of cerebral blood vessels, and perform simulation analysis to evaluate the impact of blood flow impact on plaque stability.

[0009] The multi-factor risk prediction module is used to integrate genomic data, metabolomics data and simulation results of cerebrovascular digital twins, and combine machine learning to build a multi-factor risk prediction model to identify abnormal risk characteristics and improve the accuracy of high-risk population location assessment.

[0010] The screening risk assessment engine combines the risk prediction results of the multi-factor risk prediction model with the simulation analysis results to analyze the collected multimodal data, assess the disease risk of stroke patients, screen potential high-risk groups, and achieve early screening and risk assessment of stroke.

[0011] The early warning and intervention module is used to notify patients and their families via SMS, WeChat, and email based on risk assessment results, and to provide personalized intervention plan recommendations. At the same time, it relies on the telemedicine collaboration network to conduct auxiliary diagnosis and personalized intervention treatment for patients.

[0012] A further improvement to the technical solution of the present invention is that the multimodal data acquisition and integration module specifically includes:

[0013] Multimodal data related to stroke are collected simultaneously from multiple channels, covering clinical data, imaging data, omics data, and end-point health data;

[0014] The collected multimodal data is preprocessed, including cleaning, denoising, and standardization. The preprocessed multimodal data is then integrated, and different types of data are associated and fused according to a unified format and structure to eliminate differences and redundancy between data, thus constructing a complete, standardized, and high-quality basic dataset for stroke.

[0015] A further improvement of the technical solution of the present invention is that the digital twin simulation module includes a brain blood vessel digital twin construction engine and a simulation evaluation unit;

[0016] The cerebral blood vessel digital twin construction engine is used to construct a cerebral blood vessel digital twin of a stroke patient using CT / MRI images, simulating the cerebral blood vessel structure and hemodynamic characteristics.

[0017] The simulation evaluation unit is used to simulate the impact force under different blood flow conditions on the cerebral vascular digital twin, evaluate its impact on plaque stability, quantify the risk of plaque rupture, and identify potential plaque rupture risks in advance through simulation analysis.

[0018] A further improvement to the technical solution of this invention lies in that: the brain blood vessel digital twin construction engine specifically includes:

[0019] The system automatically imports patient CT / MRI image data via the DICOM protocol, uses a multimodal registration algorithm to spatially align the image data, eliminates geometric deviations caused by different devices or scanning parameters, and accurately separates the cerebral vascular tree structure through image segmentation algorithms to extract vascular geometric features, including vascular centerline, branch points and diameter information, and constructs a vascular geometric feature database.

[0020] Based on the processed image data and the separated cerebral blood vessel tree structure, a three-dimensional digital model of cerebral blood vessels was constructed using computer graphics and medical modeling techniques. The extracted vascular geometric features were then mapped onto the three-dimensional digital model of cerebral blood vessels to restore the anatomical structure of cerebral blood vessels.

[0021] By combining computational fluid dynamics (CFD) methods to construct a hemodynamic simulation environment, and integrating specific physiological parameters of patients, including blood pressure and blood viscosity, as boundary conditions, the Navier-Stokes equations are solved through finite element analysis to simulate the velocity, pressure, and shear stress distribution of blood flow, generating dynamic blood flow field data. Then, the three-dimensional digital model of cerebral blood vessels is fused with the generated dynamic blood flow field data to form a digital twin of cerebral blood vessels for stroke patients.

[0022] A further improvement to the technical solution of the present invention is that the simulation evaluation unit specifically includes:

[0023] Based on the patient's specific physiological parameters and vascular geometric characteristics, multiple sets of blood flow simulation boundary conditions are defined in the cerebral vascular digital twin, including inlet boundary, outlet boundary and blood physical property parameters, covering typical scenarios of normal physiology, hypertensive emergencies and exercise stress.

[0024] Hemodynamic simulations were run, and the Navier-Stokes equations were solved by finite element analysis to simulate the blood flow impact force in cerebral blood vessels under different blood flow conditions. The dynamic force of the blood flow impact force on plaques on the vessel wall was analyzed. The focus was on capturing high stress concentration areas in turbulent regions, vessel bifurcation, and plaque surfaces. The wall shear stress was calculated simultaneously, the spatiotemporal fluctuation characteristics of instantaneous blood flow impact force were quantified, and a full-field stress distribution cloud map was generated.

[0025] By integrating plaque morphological parameters, including fibrous cap thickness and lipid volume, with simulated blood flow impact forces, vulnerability grading criteria are determined. The integrity of plaque structure is assessed through stress-strain analysis, and the integrity of plaque structure is analyzed in conjunction with a preset critical stress threshold. In this way, a plaque vulnerability index is output to identify potential rupture risks and help to detect and intervene in high-risk plaques in advance.

[0026] A further improvement to the technical solution of this invention lies in the following: the process of outputting the patch vulnerability index to determine the potential breakage risk is as follows:

[0027] Integrated intravascular ultrasound (IVUS) imaging was used to segment plaque fibrous caps and lipid cores, and to measure the thickness of the fibrous cap. and lipid volume percentage To obtain plaque morphological parameters and, in conjunction with the plaque fibrous cap and lipid core, determine vulnerability grading criteria, among which, thin fibrous cap... And large lipid core Defined as high-risk plaque; thick fibrous cap And small lipid core Defined as a stable plaque;

[0028] The tensile stress of the fiber cap was calculated using stress-strain analysis, taking into account the instantaneous blood flow impact force. Converted into equivalent tensile stress The critical stress threshold of the fiber cap was determined by combining the results with experiments in materials mechanics. To determine whether rupture has occurred, in order to analyze the integrity of the plaque structure, if... If it is, then it is marked as a rupture event. ;

[0029] By combining the maximum stress, critical stress, maximum shear stress gradient, reference wall shear stress value, and lipid volume experienced by the plaque during the cardiac cycle, a plaque vulnerability index is calculated using weighted averages. This index quantifies the probability of rupture risk, and risk classification thresholds are determined based on the plaque vulnerability index value to classify risk levels from 1 to 5. Level 1 represents very low risk, level 2 represents low risk, level 3 represents medium risk, level 4 represents high risk, and level 5 represents very high risk. This generates a risk heatmap, which is overlaid on a three-dimensional digital model of cerebral blood vessels to display the regions corresponding to high-risk and very high-risk levels, helping doctors or researchers identify areas in blood vessels where plaque rupture may occur.

[0030] A further improvement of the technical solution of the present invention is that the multi-factor risk prediction module includes a feature extraction and selection unit and a model construction unit;

[0031] The feature extraction and selection unit is used to extract risk features associated with risk prediction from genomic data, metabolomics data, and simulation results of cerebrovascular digital twins to form a stroke risk feature set, thereby improving the prediction accuracy and efficiency of the model.

[0032] The model building unit is used to employ a graph neural network machine learning algorithm to fuse selected risk features, construct a multi-factor risk prediction model, and output the risk prediction results of the patient's stroke in order to identify any abnormal risk features.

[0033] A further improvement to the technical solution of the present invention is that the feature extraction and selection unit specifically includes:

[0034] Feature analysis was performed on genomic data, metabolomics data, and simulation results of cerebrovascular digital twins. Signal processing techniques were used to extract risk features associated with risk prediction. Specifically, for genomic data, risk features including the MTHFR gene TT variant and the RNF213 gene variant were extracted. For metabolomics data, risk features including serum myristic acid level, methionine sulfoxide level, and the ratio of free cholesterol to cholesterol esters in very low density lipoprotein cholesterol (VLDL) were extracted. For the simulation results of cerebrovascular digital twins, risk features including wall shear stress, blood flow velocity, and blood flow rate were extracted.

[0035] Based on clinical research and simulation experiment data, corresponding allowable thresholds are pre-set for each risk characteristic and matched with the corresponding risk characteristics to form a stroke risk characteristic set, so as to comprehensively characterize the multi-dimensional information related to stroke risk.

[0036] A further improvement to the technical solution of the present invention is that the model building unit specifically includes:

[0037] Data containing various risk characteristics of patients is collected, cleaned to remove noise and missing values, and standardized one-hot encoding is used to uniformly process different types of risk characteristics, transforming them into a format suitable for graph neural network input.

[0038] A graph structure is constructed based on the correlation between risk features, with risk features as nodes and correlations as edges. The processed risk features are integrated into the node attributes. A machine learning algorithm based on graph neural networks is selected as the model architecture. The multi-factor risk prediction model is trained with graph structure data as input. The parameters are adjusted so that the multi-factor risk prediction model learns the intrinsic relationship between risk features and stroke.

[0039] After the relevant risk characteristics of new patients are processed and encoded in the same way, they are substituted into the pre-trained multi-factor risk prediction model. The model analyzes the relationship between the learned risk characteristics and stroke, compares each risk characteristic with the corresponding allowable threshold, identifies risk characteristics that exceed the limit, and then outputs the stroke risk prediction result for the patient, i.e. the comparison result of abnormal risk characteristics.

[0040] A further improvement to the technical solution of this invention lies in that: the screening risk assessment engine specifically includes:

[0041] Multimodal data, including clinical data, imaging data, omics data, and end-point health data, is extracted from the multimodal data acquisition and integration module, and the multimodal data is converted into a format suitable for input to the multifactor risk prediction model;

[0042] The risk prediction results are obtained by using a multi-factor risk prediction model, and the simulation analysis results of the cerebrovascular digital twin are called in. The two are deeply integrated to further assess the disease risk of stroke patients. Based on the risk prediction results and simulation analysis results, a comprehensive risk assessment report is generated.

[0043] Based on the risk assessment report, potential high-risk individuals are screened and marked. The marking results are then pushed to the early warning and intervention module and the stroke prevention and control management platform for intervention and reminders. Potential high-risk individuals include those with risk level 4, risk level 5, and those with at least two abnormal risk characteristics.

[0044] Due to the adoption of the above technical solution, the technical progress achieved by this invention compared to the prior art is as follows:

[0045] 1. This invention provides a stroke high-risk population localization and assessment system based on multimodal data. By integrating clinical data, imaging data, omics data, and end-point health data, it can comprehensively capture multi-dimensional information related to stroke. In particular, by using a cerebrovascular digital twin to simulate hemodynamic characteristics and combining graph neural networks to fuse multiple data sources, it can more accurately identify high-risk populations, significantly improve the accuracy and reliability of risk prediction, overcome the problem of missed detection caused by relying on a single data source in traditional methods, and provide stronger support for early intervention.

[0046] 2. This invention provides a stroke high-risk population localization and assessment system based on multimodal data. It uses CT / MRI images to construct a digital twin of cerebral blood vessels, simulates the structure and hemodynamic characteristics of cerebral blood vessels, and through simulation analysis, can detect potential plaque rupture risks in advance and quantify the risk of plaque rupture, providing a scientific basis for the early identification of high-risk populations and significantly improving the accuracy of assessment. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0048] Figure 1 This is a schematic diagram of the workflow of a stroke high-risk population location and assessment system based on multimodal data according to the present invention;

[0049] Figure 2 This is a data flow diagram of a stroke high-risk population localization and assessment system based on multimodal data according to the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] Example 1, such as Figure 1 , Figure 2 As shown, this invention provides a stroke high-risk population localization and assessment system based on multimodal data, including a stroke prevention and control management platform. The stroke prevention and control management platform has the following communication modules, wherein:

[0052] The multimodal data acquisition and integration module is used to collect stroke-related multimodal data from multiple channels, including clinical data (electronic medical records, laboratory test results), imaging data (CT / MRI images), omics data (genomics, metabolomics data), and end-point health data (physiological indicators monitored by wearable devices). It performs preprocessing operations such as cleaning, denoising, and standardization on the collected multimodal data, integrating it into a unified format stroke baseline dataset to improve data quality and eliminate data variability and redundancy. The module simultaneously collects stroke-related multimodal data from multiple channels, covering clinical data, imaging data, omics data, and end-point health data. Clinical data, such as electronic medical records and laboratory test results, is obtained through hospital information systems; CT / MRI images are collected using medical imaging equipment. Data; Genomic and metabolomics data are acquired using professional omics technologies; wearable devices are used to monitor and collect end-point health data of physiological indicators in real time, ensuring the diversity and integrity of data sources; preprocessing of the collected multimodal data includes cleaning, denoising, and standardization; data cleaning techniques are used to remove invalid, erroneous, and duplicate data; denoising algorithms are used to eliminate noise interference in the data; and data is standardized according to unified standards to ensure consistency and comparability of data from different sources, formats, and ranges, making them conform to consistent specifications and improving data quality. Then, the preprocessed multimodal data is integrated, and different types of data are associated and fused according to a unified format and structure to eliminate differences and redundancy between data, constructing a complete, standardized, and high-quality basic dataset for stroke.

[0053] The specific tasks of the multimodal data acquisition and integration module are as follows: Based on the needs of stroke research, a data acquisition path covering four categories of data—clinical, imaging, omics, and end-stage health—is planned to collect stroke-related multimodal data, including clinical data, imaging data, omics data, and end-stage health data. Clinical data is extracted in a structured manner through the hospital information system interface to ensure the complete capture of key information from electronic medical records and laboratory test results. Imaging data relies on the medical imaging equipment network and uses standardized protocols to achieve automated acquisition and transmission of CT / MRI images. Omics data is obtained through a professional laboratory platform using gene sequencing and mass spectrometry analysis technologies to acquire genomic and metabolomics data. End-stage health data is obtained through wearable devices to achieve real-time dynamic monitoring of physiological indicators such as heart rate and blood pressure. Through multi-channel data acquisition, the diversity and completeness of data sources are ensured. After data acquisition, the raw multimodal data undergoes preprocessing operations, including cleaning, denoising, and standardization. The data cleaning stage uses a rule engine and machine learning algorithms to automatically identify and remove missing and outlier values. The system includes several steps: First, it establishes a data integrity verification mechanism to address duplicate records. Second, in the denoising stage, filtering algorithms and statistical correction methods are used to eliminate errors caused by equipment noise and environmental interference, tailored to the characteristics of different modalities. Third, in the standardization stage, numerical data is normalized according to international medical data standards, and a unified coding system is established for categorical data to ensure consistency in dimensions, format, and semantics across different sources, resulting in a clean dataset that meets analytical requirements. Fourth, it integrates preprocessed multimodal data, establishes cross-modal data mapping relationships, aligns clinical data with imaging and omics data spatiotemporally, constructs a patient lifecycle data map, and defines data meta-standards using ontology methods to eliminate terminology differences between different systems, achieving semantic interoperability between structured and semi-structured data. Finally, through data warehouse technology, multimodal data is stored in association while protecting data privacy, forming a complete, standardized, and high-quality basic stroke dataset. The integrated basic stroke dataset is scalable, supports dynamic updates and version management, and a multi-level access control mechanism is established to ensure data security and compliance.

[0054] The digital twin simulation module is used to construct a digital twin of the cerebral blood vessels of stroke patients, simulate the structure and hemodynamic characteristics of cerebral blood vessels, and perform simulation analysis to evaluate the impact of blood flow impact on plaque stability. The digital twin simulation module includes a cerebral blood vessel digital twin construction engine and a simulation evaluation unit.

[0055] Among them, the cerebral vascular digital twin construction engine is used to construct digital twins of the cerebral blood vessels of stroke patients using CT / MRI images, simulating the structure and hemodynamic characteristics of cerebral blood vessels. It automatically imports the patient's CT / MRI image data via the DICOM protocol, uses a multimodal registration algorithm to spatially align the image data, eliminates geometric deviations caused by different devices or scanning parameters, and accurately separates the cerebral vascular tree structure using image segmentation algorithms. It extracts vascular geometric features, including the vessel centerline, branch points, and diameter information, and constructs a vascular geometric feature database. Based on the processed image data and the separated cerebral vascular tree structure, it utilizes computation... Using computer graphics and medical modeling techniques, a three-dimensional digital model of cerebral blood vessels is constructed, and the extracted vascular geometric features are mapped onto the three-dimensional digital model of cerebral blood vessels to restore the anatomical structure of cerebral blood vessels. Combining computational fluid dynamics (CFD) methods, a hemodynamic simulation environment is constructed, integrating specific physiological parameters of patients, including blood pressure and blood viscosity, as boundary conditions. The Navier-Stokes equations are solved through finite element analysis to simulate the velocity, pressure, and shear stress distribution of blood flow, generating dynamic blood flow field data. Then, the three-dimensional digital model of cerebral blood vessels is fused with the generated dynamic blood flow field data to form a digital twin of the cerebral blood vessels of stroke patients.

[0056] The specific functions of the cerebral vascular digital twin construction engine are as follows: The engine automatically connects to medical imaging equipment via the DICOM protocol, parses patient information (ID, age, gender), scanning parameters (slice thickness, resolution, reconstruction kernel), and equipment model from the DICOM tags, constructs a structured image metadata database, converts the DICOM file to NIfTI format (Neuroimaging Informatics Technology Initiative standard), achieves lossless import and standardized storage of patient CT / MRI image data, and employs a multimodal registration algorithm based on rigid / non-rigid transformation models (rigid transformation models use mutual information algorithms to perform global spatial alignment of CT / MRI images of the same patient at different time phases, eliminating scanning posture differences; non-rigid transformation models use B-spline free deformation algorithms to correct local geometric deviations caused by brain tissue deformation, ensuring pixel-level correspondence of multimodal images). This spatial alignment eliminates geometric deviations and ensures the spatiotemporal consistency of multi-temporal or multimodal image data, thereby enabling... By combining image segmentation algorithms with vascular enhancement filtering techniques, the cerebral vascular tree structure is accurately separated from aligned image data. Vascular geometric features, including the vascular centerline, branch points, and diameter, are extracted. Simultaneously, vascular connectivity is verified through topological analysis, and a vascular geometric feature database is constructed to ensure the integrity and accuracy of cerebral vascular anatomy details. The vascular enhancement filtering technique consists of multi-scale Hessian matrix analysis and Frangi filter optimization. For multi-scale Hessian matrix analysis, eigenvalues ​​of the local Hessian matrix are calculated to identify vascular structures (tubular features) and suppress background noise and non-vascular tissue. For Frangi filter optimization, prior knowledge of vascular diameter is used to dynamically adjust the filter scale parameters, enhancing the contrast of vessels of different thicknesses. For vascular centerline extraction, distance transformation and skeletonization algorithms are used to generate the vascular centerline and calculate the branch point coordinates and topological relationships. For diameter measurement, cross-sections of the vessel are sampled along the centerline normal direction, and the local diameter is calculated through ellipse fitting to generate a diameter variation curve.After obtaining the vascular geometric features of the cerebral vascular tree structure, a surface reconstruction algorithm from computer graphics (including implicit surface representation and subdivision surface optimization) was employed. Through implicit surface representation, based on radial basis function (RBF) interpolation, the vascular centerline and diameter information were transformed into continuous surface equations, avoiding the discretization error of explicit meshes. Subdivision surface optimization, using the Loop subdivision algorithm, iteratively smoothed the initial mesh, eliminating the staircase effect and ensuring a smooth model surface. Combined with medical modeling techniques, the vascular centerline and diameter information were transformed into a three-dimensional digital model of the cerebral blood vessels. Subdivision surface optimization and smoothing further eliminated geometric discretization errors and restored the original vascular structure. The continuous anatomical morphology of cerebral blood vessels was analyzed. Simultaneously, extracted vascular geometric features were mapped onto a three-dimensional digital model of cerebral blood vessels, constructing a digital vascular network including topological relationships and morphological parameters. This reconstructed the anatomical structure of cerebral blood vessels, including their shape and size, as well as the spatial relationships and connections between them. Computational fluid dynamics (CFD) methods were integrated, and simulation boundary conditions were set based on patient-specific physiological parameters such as blood pressure and blood viscosity. These conditions included inlet and outlet boundaries, and blood physical properties. For the inlet boundary, an inlet pressure waveform (pulsating sine function) was set based on patient-specific blood pressure data. For the outlet boundary, a Windkessel model (three elements) was used. A resistance-capacitance model was used to simulate peripheral vascular resistance. Parameters were estimated based on patient age and gender. For blood physical properties, the Newtonian fluid viscosity coefficient (μ = 3.5 × 10⁻³ Pa·s, dynamically adjusted within ±20%) was adjusted based on the patient's blood viscosity (measured using a rotational viscometer) and hematocrit (HCT). This constructed a hemodynamic simulation environment. Steady-state / transient analyses were performed by solving the Navier-Stokes equations using finite element analysis. For steady-state analysis, the average blood flow velocity and pressure distribution were calculated to assess the degree of vascular stenosis (stenosis rate = (1 - minimum cross-sectional area / reference cross-sectional area) × 100%). For transient simulation, P... The ISO algorithm (pressure implicit segmentation operator) is used to solve unsteady flow with a time step of 0.001s to simulate the dynamic changes in blood flow within one cardiac cycle (0.8s). At the same time, a turbulence model is selected to simulate the laminar / turbulent state of blood in cerebral blood vessels (the turbulence model is automatically switched according to the Reynolds number (Re<2300 uses laminar flow, Re≥4000 uses the k-ε turbulence model, and the intermediate transition zone uses the low Reynolds number k-ω model)). Dynamic blood flow field data is generated, including velocity, pressure and shear stress distribution. Then, the three-dimensional digital model of cerebral blood vessels is fused with the generated dynamic blood flow field data to generate an interactive digital twin of the cerebral blood vessels of stroke patients.

[0057] The simulation evaluation unit is used to simulate the impact forces under different blood flow conditions on a cerebral vascular digital twin, assess its impact on plaque stability, quantify the risk of plaque rupture, and identify potential plaque rupture risks in advance through simulation analysis. Based on patient-specific physiological parameters and vascular geometry, multiple sets of blood flow simulation boundary conditions are defined in the cerebral vascular digital twin, including inlet and outlet boundaries and blood physical property parameters, covering typical scenarios such as normal physiology, hypertensive emergencies, and exercise stress. Hemodynamic simulations are run, and the Navier-Stokes equations are solved using finite element analysis to simulate blood flow impacts within cerebral vessels under different blood flow conditions. The system analyzes the dynamic force of blood flow impact on plaques on the vessel wall, focusing on capturing high-stress concentration areas in turbulent regions, vessel bifurcation, and plaque surfaces. It simultaneously calculates wall shear stress, quantifies the spatiotemporal fluctuation characteristics of instantaneous blood flow impact, generates a full-field stress distribution cloud map, integrates plaque morphological parameters including fibrous cap thickness and lipid volume with simulated blood flow impact, determines vulnerability grading criteria, assesses plaque structural integrity through stress-strain analysis, analyzes plaque structural integrity in conjunction with preset critical stress thresholds, and outputs a plaque vulnerability index to identify potential rupture risks, helping to detect and intervene in high-risk plaques in advance.

[0058] The specific tasks of the simulation evaluation unit are as follows: Based on patient-specific physiological parameters and vascular geometry, construct blood flow simulation boundary conditions covering typical scenarios such as normal physiology, hypertensive emergencies, and exercise stress. These boundary conditions include inlet and outlet boundaries, and blood physical properties. The inlet boundary uses dynamic blood pressure monitoring data to set the pressure waveform and adjusts the pulse frequency based on heart rate (normal 60-100 bpm, hypertensive emergencies >120 bpm) to simulate different blood flow drive modes. The outlet boundary uses the Windkessel model, adjusting model parameters (resistance coefficient, compliance) according to patient age, vascular elasticity, and peripheral resistance to reflect the differences in peripheral circulatory load under different physiological states. Blood physical properties (viscosity, density) are dynamically adjusted based on the patient's blood viscosity and hematocrit, covering normal viscosity (3.5-4.5 mPa·s) and increased blood viscosity in hypertension (>5.0 mPa·s). In scenarios involving blood dilution (mPa·s) and post-exercise blood dilution (viscosity reduction of 10%-15%), blood density is dynamically adjusted based on hematocrit, which is obtained from laboratory test results to ensure a high degree of consistency between the simulated blood flow boundary conditions and clinical reality. Finite element analysis is used to solve the Navier-Stokes equations to simulate hemodynamic behavior within cerebral blood vessels under different simulated blood flow boundary conditions. The focus is on capturing high stress concentration phenomena in turbulent regions, vascular bifurcation, and plaque surfaces. A subgrid model in Large Eddy Simulation (LES) is used to analyze the turbulent structure, quantifying the spatiotemporal fluctuation characteristics of instantaneous blood flow impact force (force per unit area, N / m²), including peak stress, duration of action, and frequency distribution. Wall shear stress is calculated simultaneously, and a full-field stress distribution cloud map is generated through spatial interpolation to mark high-stress areas. By combining blood flow velocity vector field analysis, the dynamic loading mode of blood flow impact force on plaques is clarified, namely periodic impact or continuous shear. Based on intravascular ultrasound (IVUS) images, plaque morphological parameters including fibrous cap thickness and lipid volume are obtained, and vulnerability grading criteria are determined simultaneously. Plaque morphological parameters and blood flow impact force simulation data are integrated, and blood flow impact force is mapped to plaque structure through stress-strain analysis. The tensile stress and strain of fibrous cap are calculated. Combined with the critical stress threshold determined by material mechanics experiments, the integrity of plaque structure is determined, and a plaque vulnerability index is introduced. By comprehensively considering stress level, stress gradient and plaque lipid volume, the probability of rupture risk is quantified and divided into risk levels of 1-5. A risk heat map is generated to show the plaque location and stress distribution characteristics corresponding to high risk level and very high risk level.

[0059] The Navier-Stokes equations are expressed as follows:

[0060] ;

[0061] In the formula: This represents the blood flow velocity vector (m / s). For time (s), Blood flow density (kg / m³) Pressure (Pa). The viscosity is kinematic (m² / s). The external force is N / kg. In cerebral blood vessels, the blood flow velocity ranges from 0.1 to 1.5 m / s, depending on the vessel diameter and blood flow state. The blood density is approximately 1050 kg / m³, and the blood pressure ranges from 10000 to 18000 Pa (75-135 mmHg). The kinematic viscosity of blood is approximately 3.5-4.5 mPa·s under normal conditions, and may exceed 5.0 mPa·s in cases of hypertension. In cerebral vascular simulation, the external force is usually ignored and assumed to be 0.

[0062] The expression for the subgrid model in Large Eddy Simulation (LES) is as follows:

[0063] ;

[0064] In the formula: Let be the subgrid stress tensor, representing the additional stress caused by turbulent motion at the subgrid scale; and This is the filtered value of blood flow velocity, i.e., the average velocity component after filtering. Filtering is used to remove small-scale turbulent fluctuations and retain only large-scale flow characteristics; is the subgrid viscosity coefficient, a parameter used to simulate turbulent viscosity at the subgrid scale; The Kronecker notation is used to indicate whether two indices are equal. hour, ;when hour, ; and The average velocity gradient represents the rate of change of velocity in space. The sum of squares of the velocity filtering values ​​represents the sum of the squares of all velocity components; the subgrid viscosity coefficient is typically related to the grid size and fluid properties, ranging from... arrive Between m² / s, the magnitude of the subgrid stress tensor depends on the local velocity gradient and the subgrid viscosity, typically within the range of m² / s. arrive Between Pa;

[0065] The expression for the spatiotemporal fluctuation characteristics of instantaneous blood flow impact force is as follows:

[0066] ;

[0067] ;

[0068] In the formula: For time The force per unit area, i.e. the instantaneous impact force of blood flow, is measured in Newtons per square meter (N / m²). Dynamic pressure, representing time. Pressure inside blood vessels, measured in millimeters of mercury (mmHg). Blood flow velocity, representing time. The speed at which blood flows is measured in meters per second (m / s). Diastolic pressure is the lowest pressure in the arteries when the heart relaxes. Systolic blood pressure is the highest pressure in the arteries when the heart contracts. The duration of the cardiac cycle is the time required for the heart to complete one contraction and relaxation. , Heart rate, measured in beats per minute. It is a sinusoidal function used to simulate the periodic pressure changes in the heart, with phase shift. Ensure that stress is at its lowest point at the beginning of the cardiac cycle;

[0069] The expression for wall shear stress is as follows:

[0070] ;

[0071] In the formula: Wall shear stress (Pa) is the force exerted by blood flow on the vessel wall and is used to assess the impact of blood flow on the vessel wall. The viscosity is the dynamic viscosity (Pa·s). This represents the blood flow velocity vector (m / s). The normal distance to the wall (m) is the wall shear stress in cerebral blood vessels, which is usually between 0.1 and 5 Pa.

[0072] Furthermore, the process of outputting the plaque fragility index to determine the potential rupture risk is as follows:

[0073] Integrated intravascular ultrasound (IVUS) imaging was used to segment plaque fibrous caps and lipid cores, and to measure the thickness of the fibrous cap. and lipid volume percentage To obtain plaque morphological parameters and, in conjunction with the plaque fibrous cap and lipid core, determine vulnerability grading criteria, among which, thin fibrous cap... And large lipid core Defined as high-risk plaque; thick fibrous cap And small lipid core Defined as a stable plaque; tensile stress of the fibrous cap is calculated using stress-strain analysis, taking into account the instantaneous blood flow impact force. Converted into equivalent tensile stress The critical stress threshold of the fiber cap was determined by combining the results with experiments in materials mechanics. To determine whether rupture has occurred, in order to analyze the integrity of the plaque structure, if... If it is, then it is marked as a rupture event. The plaque vulnerability index is calculated by weighting the maximum stress, critical stress, maximum shear stress gradient, reference wall shear stress value, and lipid volume experienced by the plaque during the cardiac cycle. This quantifies the probability of rupture risk, and risk classification thresholds are determined based on the plaque vulnerability index value to classify risk levels from 1 to 5. Level 1 is extremely low risk, level 2 is low risk, level 3 is medium risk, level 4 is high risk, and level 5 is extremely high risk. This generates a risk heat map, which is overlaid on a three-dimensional digital model of cerebral blood vessels to show the areas corresponding to high-risk and extremely high-risk levels, helping doctors or researchers identify areas in blood vessels where plaque rupture may occur.

[0074] The expression for the plaque vulnerability index is as follows:

[0075] ;

[0076] In the formula: This is a plaque vulnerability index used to assess the stability and rupture risk of arterial plaques; is a weighting coefficient. , , The values ​​are 0.5, 0.3, and 0.2, respectively, used to adjust different parameters in the calculation. The importance of time Maximum stress represents the maximum stress that the plaque experiences during a cardiac cycle. The critical stress threshold is the stress threshold at which plaque begins to fracture. The maximum shear stress gradient represents the maximum spatial rate of change of wall shear stress (WSS). The reference wall shear stress value is given as 10 Pa to normalize the shear stress gradient. This represents the lipid volume percentage, indicating the volume of the lipid core within the plaque.

[0077] Multiple risk levels correspond one-to-one with multiple risk classification thresholds, as follows:

[0078] Level 1 is an extremely low risk level, and its risk classification threshold is: ;

[0079] Level 2 is a low-risk level, and its risk classification threshold is: ;

[0080] Level 3 is a medium-risk level, and its risk classification threshold is: ;

[0081] Level 4 is a high-risk level, and its risk classification threshold is: ;

[0082] Level 5 is an extremely high risk level, and its risk classification threshold is: ;

[0083] The multi-factor risk prediction module is used to integrate genomic data, metabolomics data and simulation results of cerebrovascular digital twins, and combine machine learning to build a multi-factor risk prediction model to identify abnormal risk characteristics and improve the accuracy of high-risk population location assessment.

[0084] The screening risk assessment engine combines the risk prediction results of the multi-factor risk prediction model with the simulation analysis results to analyze the collected multimodal data, assess the disease risk of stroke patients, screen potential high-risk groups, and achieve early screening and risk assessment of stroke.

[0085] The early warning and intervention module is used to notify patients and their families via SMS, WeChat, and email based on risk assessment results, and to provide personalized intervention plan recommendations. At the same time, it relies on the telemedicine collaboration network to conduct auxiliary diagnosis and personalized intervention treatment for patients.

[0086] Example 2, as Figure 1 , Figure 2 As shown, based on Embodiment 1, the present invention provides a technical solution: the multi-factor risk prediction module includes a feature extraction and selection unit and a model building unit;

[0087] The feature extraction and selection unit is used to extract risk features associated with risk prediction from genomic data, metabolomics data, and simulation results of cerebrovascular digital twins, forming a stroke risk feature set to improve the prediction accuracy and efficiency of the model. It performs feature analysis on genomic data, metabolomics data, and simulation results of cerebrovascular digital twins, and uses signal processing technology to extract risk features associated with risk prediction. Specifically, for genomic data, it extracts risk features including MTHFR gene TT variants and RNF213 gene variants; for metabolomics data, it extracts risk features including serum myristic acid levels, methionine sulfoxide levels, and the ratio of free cholesterol to cholesterol esters in very low density lipoprotein cholesterol (VLDL); for simulation results of cerebrovascular digital twins, it extracts risk features including wall shear stress, blood flow velocity, and blood flow rate. Based on clinical research and simulation experimental data, it pre-sets corresponding allowable thresholds for each risk feature and matches them with the corresponding risk features to integrate and form a stroke risk feature set, comprehensively representing multi-dimensional information related to stroke risk.

[0088] The specific tasks of the feature extraction and selection unit are as follows: Based on the patient's stroke risk prediction needs, feature analysis is performed on genomic data, metabolomics data, and simulation results of cerebrovascular digital twins to extract risk features associated with risk prediction. Specifically, risk features from genomic data include MTHFR gene TT variants and RNF213 gene variants; risk features from metabolomics data include serum myristate levels, methionine sulfoxide levels, and the ratio of free cholesterol to cholesterol esters in very low-density lipoprotein cholesterol (VLDL); and risk features from cerebrovascular digital twin simulation results include wall shear stress. Blood flow velocity and blood volume; based on the risk characteristics of genomic data, the TT variant of the MTHFR gene leads to a reduction of enzyme activity of more than 50%, causing a surge in homocysteine ​​levels in the blood, damaging the vascular endothelium and accelerating thrombus formation. The risk of stroke is 2-3 times higher in people with the TT variant. Timely folic acid supplementation (0.8 mg / day) can reduce homocysteine ​​by 25% and reduce the risk of stroke by 18%. The RNF213 gene variant has a detection rate of 12.7% in stroke patients under the age of 55 and is strongly associated with atherosclerotic stroke of large arteries. Individuals with the variant have a significantly increased risk of large artery atherosclerosis and need to be screened for carotid artery plaques regularly. Based on the risk characteristics of metabolomics data, serum myristic acid level, a long-chain dicarboxylic acid, is closely associated with ischemic stroke events; for every 1 standard deviation increase in serum myristic acid level, the risk of ischemic stroke increases by 30% (based on data from a community atherosclerosis risk study); methionine sulfoxide level, an amino acid metabolite, is significantly associated with the risk of ischemic stroke; the stroke risk in the highest quartile of methionine sulfoxide levels is 2.1 times that of the lowest quartile; the ratio of free cholesterol to cholesterol esters in very low-density lipoprotein cholesterol (VLDL) is a lipid metabolism indicator and is positively correlated with the risk of all-cause stroke; for every 1 standard deviation increase in this ratio, the risk of stroke increases. The risk increased by 25% (based on data from 118,021 participants in the UK Biobank); regarding the risk characteristics of the simulation results of the cerebrovascular digital twin, uneven distribution of wall shear stress increases the risk of stroke; when the maximum stress of the cerebrovascular wall exceeds the normal range, the risk of stroke increases significantly; abnormal blood flow velocity and blood flow indicate the risk of stroke, and reduced blood flow velocity or blood flow in cerebrovascular vessels indicates an increased risk of stroke; based on clinical research and simulation experimental data, corresponding allowable thresholds were set for each identified risk characteristic, and each allowable threshold was mapped one-to-one with the risk characteristic, integrating multi-dimensional risk characteristics to form a stroke risk characteristic set;

[0089] The model building unit is used to construct a multi-factor risk prediction model by using a graph neural network machine learning algorithm, fusing selected risk features, and outputting the risk prediction results of stroke in patients. This involves identifying abnormal risk features, collecting data containing various risk features of patients, cleaning it to remove noise and missing values, standardizing and one-hot encoding different types of risk features to convert them into a format suitable for graph neural network input, constructing a graph structure based on the relationships between risk features (as nodes, relationships as edges), and integrating the processed risk features into node attributes. A graph neural network-based machine learning algorithm is selected as the model architecture, and the multi-factor risk prediction model is trained using graph structure data as input. Parameters are adjusted to enable the multi-factor risk prediction model to learn the intrinsic connection between risk features and stroke. The relevant risk features of new patients are processed and encoded in the same way, then substituted into the trained multi-factor risk prediction model. The model analyzes the learned relationship between risk features and stroke, compares each risk feature with its corresponding allowable threshold, identifies risk features exceeding the limit, and outputs the stroke risk prediction result for that patient, i.e., the comparison result of abnormal risk features.

[0090] The specific tasks of the model construction unit are as follows: Collecting various risk characteristic data from multiple medical databases, collaborative projects with research institutions, and clinical records, covering a large number of known stroke patients, including genomic, metabolomic, and cerebrovascular digital twin simulation results. This ensures a sufficient number of samples are collected to cover diverse stroke risk profiles, guaranteeing the reliability and generalization ability of the model training. Data cleaning is performed to remove noise and missing values, ensuring data quality. For different types of risk characteristics, standardized methods are used to process continuous data, conforming it to a specific distribution for easier subsequent analysis. Discrete data is transformed using one-hot encoding. The process transforms raw risk feature data into a computer-readable numerical form, converting it into a unified format suitable for graph neural network input. A graph structure is constructed based on the relationships between risk features, with various risk features acting as nodes. Edges in the graph structure are determined based on the relationships between risk features, which are established through clinical research, literature reviews, and expert experience. This constructs a network reflecting the relationships between risk features. Simultaneously, the processed risk features are integrated into node attributes to enrich node information. Based on this, a machine learning algorithm based on graph neural networks (specifically, a graph convolutional network (GCN)) is used to build the model architecture, including an input layer, hidden layers, and so on. The model consists of hidden and output layers. The input layer takes graph structure data, including nodes and edges. The hidden layer learns the relationships and feature representations between nodes through graph convolutional layers or graph attention layers. The output layer outputs the risk prediction result for each node. Using the constructed graph structure data as input, a multi-factor risk prediction model is trained. The graph structure data contains node information (risk features) and edge information (relationships). During training, model parameters, including the learning rate and number of iterations, are continuously adjusted to enable the multi-factor risk prediction model to learn the intrinsic connection between risk features and stroke, improving the model's accuracy and reliability in risk prediction. When a new patient needs stroke treatment... When predicting intermediate risk, the relevant risk characteristics of new patients are processed and coded in the same way as the training data to ensure data format consistency. Then, the processed risk characteristic data is substituted into the pre-trained multi-factor risk prediction model. Based on the intrinsic relationship between risk characteristics and stroke learned during the learning phase, the data of new patients is analyzed, and the processed risk characteristics are compared with the preset allowable thresholds for each risk characteristic. Risk characteristics that do not meet the allowable threshold requirements are identified as abnormal risk characteristics. The risk prediction result of stroke for the patient is then output, which includes abnormal risk characteristics and corresponding allowable thresholds.

[0091] The screening risk assessment engine specifically includes: extracting multimodal data, including clinical data, imaging data, omics data, and end-point health data, from the multimodal data acquisition and integration module; converting the multimodal data into a format suitable for input to the multifactor risk prediction model; using the multifactor risk prediction model to obtain risk prediction results; and simultaneously calling the simulation analysis results of the cerebrovascular digital twin to deeply integrate the two to further assess the disease risk of stroke patients. Based on the risk prediction results and simulation analysis results, a comprehensive risk assessment report is generated. According to the risk assessment report, potential high-risk groups are screened and marked. The marking results are pushed to the early warning and intervention module and the stroke prevention and control management platform for intervention and reminders. Among them, potential high-risk groups include risk level 4, risk level 5, and risk characteristics with at least two abnormalities at the same time.

[0092] The specific functions of the screening risk assessment engine are as follows: It extracts clinical data, imaging data, omics data, and end-point health data from the multimodal data acquisition and integration module. This encompasses comprehensive patient health information, including electronic medical records, laboratory test results, CT / MRI images, genomic and metabolomics data, and physiological indicators monitored by wearable devices. The multimodal data is then standardized and format-converted to ensure consistency and compatibility, meeting the input requirements of the multifactor risk prediction model and eliminating data variability and redundancy. After data preprocessing, the multifactor risk prediction model analyzes the processed multimodal data to obtain preliminary predictions of stroke risk and identify any abnormal risk characteristics. Simultaneously, it integrates the simulation analysis results of a cerebrovascular digital twin, deeply fusing the two. By integrating the quantitative analysis of the risk prediction model with the simulation of the digital twin, it achieves a comprehensive assessment of the patient's health. Further assessment of stroke patients' disease risk involves analyzing their actual risk from different dimensions. This includes not only static risk factors but also dynamic physiological states and vascular hemodynamic characteristics, providing a more comprehensive and accurate risk assessment. This generates a comprehensive risk assessment report, providing a basis for subsequent intervention measures. Based on the generated risk assessment report, a screening engine identifies and marks potential high-risk individuals. These individuals are those identified as having a higher risk of stroke. Potential high-risk individuals include those with risk levels 4 and 5, as well as those with at least two abnormal risk characteristics simultaneously, requiring timely intervention and management. The marking results are then pushed to the early warning and intervention module and the stroke prevention and control management platform for targeted intervention reminders. This enables early screening, accurate assessment, and timely intervention of stroke, reducing the incidence and mortality of stroke.

[0093] The early warning and intervention module specifically includes: receiving risk assessment reports and labeling results of potentially high-risk individuals generated by the screening risk assessment engine; performing in-depth analysis of the risk assessment reports and labeling results to determine the patient's basic information, risk level, and abnormal risk characteristics; and immediately initiating subsequent notification and intervention processes to ensure no critical opportunities are missed. After identifying patients requiring intervention and their corresponding risk characteristics, risk warning notifications are simultaneously sent to the patient and their family via three mainstream communication methods: SMS, WeChat, and email. The notification content is concise and clear, covering the risk level and abnormal risk characteristics. Furthermore, based on the patient's risk characteristics and health status data, personalized intervention plan recommendations are generated. The guidelines include dietary adjustments, exercise recommendations, and medical advice, ensuring they are targeted and practical. This helps patients and their families clearly understand the coping measures and actively participate in disease prevention and control. After the notification and intervention plan are issued, the patient's risk assessment report, relevant examination data, and personalized intervention plan information are shared in real time with the collaborating medical expert team through a telemedicine network. The patient undergoes further auxiliary diagnosis through a remote consultation platform, and the personalized intervention plan is optimized and improved based on feedback from on-site medical personnel. If the patient's condition requires it, personalized intervention treatment under remote guidance can be arranged to ensure that the patient can receive timely and efficient medical services and achieve effective prevention and control of stroke.

[0094] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A stroke high-risk population localization and assessment system based on multimodal data, comprising a stroke prevention and control management platform, characterized in that, The stroke prevention and management platform has the following communication connection modules, including: The multimodal data acquisition and integration module is used to collect multimodal data related to stroke from multiple channels, and to preprocess the collected multimodal data to integrate it into a stroke basic dataset in a unified format. The digital twin simulation module is used to construct a digital twin of the cerebral blood vessels of stroke patients, simulate the structure and hemodynamic characteristics of cerebral blood vessels, and perform simulation analysis to evaluate the impact of blood flow impact on plaque stability. The digital twin simulation module includes a cerebral blood vessel digital twin construction engine and a simulation evaluation unit. The cerebral vascular digital twin construction engine is used to construct a cerebral vascular digital twin of a stroke patient using CT / MRI images, simulating the cerebral vascular structure and hemodynamic characteristics, specifically including: The patient's CT / MRI image data is automatically imported via the DICOM protocol. The image data is spatially aligned using a multimodal registration algorithm. The cerebral vascular tree structure is accurately separated using an image segmentation algorithm. Vascular geometric features, including vascular centerline, branch points, and diameter information, are extracted to construct a vascular geometric feature database. Based on the processed image data and the separated cerebral blood vessel tree structure, a three-dimensional digital model of cerebral blood vessels was constructed using computer graphics and medical modeling techniques. The extracted vascular geometric features were then mapped onto the three-dimensional digital model of cerebral blood vessels to restore the anatomical structure of cerebral blood vessels. By combining computational fluid dynamics methods to construct a hemodynamic simulation environment, and integrating specific physiological parameters of patients, including blood pressure and blood viscosity, as boundary conditions, the Navier-Stokes equations are solved through finite element analysis to simulate the velocity, pressure, and shear stress distribution of blood flow, generating dynamic blood flow field data. Then, the three-dimensional digital model of cerebral blood vessels is fused with the generated dynamic blood flow field data to form a digital twin of cerebral blood vessels for stroke patients. The simulation evaluation unit is used to simulate the impact force under different blood flow conditions on a cerebral vascular digital twin, evaluate its impact on plaque stability, quantify the risk of plaque rupture, and identify potential plaque rupture risks in advance through simulation analysis, specifically including: Based on the patient's specific physiological parameters and vascular geometric characteristics, multiple sets of blood flow simulation boundary conditions are defined in the cerebral vascular digital twin, including inlet boundary, outlet boundary and blood physical property parameters, covering typical scenarios of normal physiology, hypertensive emergencies and exercise stress. Hemodynamic simulations were run, and the Navier-Stokes equations were solved by finite element analysis to simulate the blood flow impact force in cerebral blood vessels under different blood flow conditions. The dynamic force of the blood flow impact force on plaques on the vessel wall was analyzed. The focus was on capturing high stress concentration areas in turbulent regions, vessel bifurcation, and plaque surfaces. The wall shear stress was calculated simultaneously, the spatiotemporal fluctuation characteristics of instantaneous blood flow impact force were quantified, and a full-field stress distribution cloud map was generated. By integrating plaque morphological parameters, including fibrous cap thickness and lipid volume, with simulated blood flow impact force, vulnerability grading criteria are determined. The integrity of plaque structure is assessed through stress-strain analysis, and the integrity of plaque structure is analyzed in conjunction with a preset critical stress threshold. Finally, a plaque vulnerability index is output to determine the potential rupture risk. The multi-factor risk prediction module is used to integrate genomic data, metabolomics data and simulation results of cerebrovascular digital twins, and combine machine learning to build a multi-factor risk prediction model to identify abnormal risk characteristics. The screening risk assessment engine combines the risk prediction results of a multi-factor risk prediction model with simulation analysis results to screen potential high-risk groups. The early warning and intervention module is used to notify patients and their families based on risk assessment results and to provide personalized intervention recommendations.

2. The stroke high-risk population localization and assessment system based on multimodal data according to claim 1, characterized in that: The multimodal data acquisition and integration module specifically includes: Multimodal data related to stroke are collected simultaneously from multiple channels, covering clinical data, imaging data, omics data, and end-point health data; The collected multimodal data is preprocessed, including cleaning, denoising, and standardization, and then the preprocessed multimodal data is integrated to construct a basic dataset for stroke.

3. The stroke high-risk population localization and assessment system based on multimodal data according to claim 1, characterized in that: The process of determining potential rupture risk by outputting the patch fragility index is as follows: Integrated intravascular ultrasound imaging segmentation of plaque fibrous cap and lipid core, and measurement of fibrous cap thickness. and lipid volume percentage To obtain plaque morphological parameters and, in conjunction with the plaque fibrous cap and lipid core, determine vulnerability grading criteria, among which, thin fibrous cap... And large lipid core Defined as high-risk plaque; thick fibrous cap And small lipid core Defined as a stable plaque; The tensile stress of the fiber cap was calculated using stress-strain analysis, taking into account the instantaneous blood flow impact force. Converted into equivalent tensile stress The critical stress threshold of the fiber cap was determined by combining the results with experiments in materials mechanics. To determine whether rupture has occurred, in order to analyze the integrity of the plaque structure, if... If it is, then it is marked as a rupture event. ; The plaque vulnerability index is calculated by weighting the maximum stress, critical stress, maximum shear stress gradient, reference wall shear stress value, and lipid volume experienced by the plaque during the cardiac cycle. This quantifies the probability of rupture risk. Based on the value of the plaque vulnerability index, a risk classification threshold is determined to divide the risk into 5 levels: level 1 is extremely low risk, level 2 is low risk, level 3 is medium risk, level 4 is high risk, and level 5 is extremely high risk. A risk heat map is then generated and overlaid on a three-dimensional digital model of cerebral blood vessels to display the regions corresponding to high-risk and extremely high-risk levels.

4. The stroke high-risk population localization and assessment system based on multimodal data according to claim 1, characterized in that: The multi-factor risk prediction module includes a feature extraction and selection unit and a model building unit; The feature extraction and selection unit is used to extract risk features associated with risk prediction from genomic data, metabolomics data and simulation results of cerebrovascular digital twins to form a stroke risk feature set. The model building unit is used to employ a graph neural network machine learning algorithm to fuse selected risk features, construct a multi-factor risk prediction model, and output the risk prediction results of the patient's stroke in order to identify any abnormal risk features.

5. A stroke high-risk population localization and assessment system based on multimodal data according to claim 4, characterized in that: The feature extraction and selection unit specifically includes: Feature analysis was performed on genomic data, metabolomics data, and simulation results of cerebrovascular digital twins. Signal processing techniques were used to extract risk features associated with risk prediction. Specifically, for genomic data, risk features including MTHFR gene TT variant and RNF213 gene variant were extracted; for metabolomics data, risk features including serum myristic acid level, methionine sulfoxide level, and the ratio of free cholesterol to cholesterol ester in very low density lipoprotein cholesterol were extracted; and for cerebrovascular digital twin simulation results, risk features including wall shear stress, blood flow velocity, and blood flow rate were extracted. Based on clinical research and simulation experiment data, corresponding allowable thresholds are pre-set for each risk characteristic and matched with the corresponding risk characteristics to form a stroke risk characteristic set.

6. The stroke high-risk population localization and assessment system based on multimodal data according to claim 5, characterized in that: The model building unit specifically includes: Data containing various risk characteristics of patients is collected, cleaned to remove noise and missing values, and standardized one-hot encoding is used to uniformly process different types of risk characteristics, transforming them into a format suitable for graph neural network input. A graph structure is constructed based on the correlation between risk features, with risk features as nodes and correlations as edges. The processed risk features are integrated into the node attributes. A machine learning algorithm based on graph neural networks is selected as the model architecture. The multi-factor risk prediction model is trained with graph structure data as input. The parameters are adjusted so that the multi-factor risk prediction model learns the intrinsic relationship between risk features and stroke. After the relevant risk characteristics of new patients are processed and encoded in the same way, they are substituted into the pre-trained multi-factor risk prediction model. The model analyzes the relationship between the learned risk characteristics and stroke, compares each risk characteristic with the corresponding allowable threshold, identifies risk characteristics that exceed the limit, and then outputs the stroke risk prediction result for the patient, i.e. the comparison result of abnormal risk characteristics.

7. A stroke high-risk population localization and assessment system based on multimodal data according to claim 6, characterized in that: The screening risk assessment engine specifically includes: Multimodal data, including clinical data, imaging data, omics data, and end-point health data, is extracted from the multimodal data acquisition and integration module, and the multimodal data is converted into a format suitable for input to the multifactor risk prediction model; The risk prediction results are obtained by using a multi-factor risk prediction model, and the simulation analysis results of the cerebrovascular digital twin are called in. The two are deeply integrated to further assess the disease risk of stroke patients. Based on the risk prediction results and simulation analysis results, a comprehensive risk assessment report is generated. Based on the risk assessment report, potential high-risk individuals are screened and marked. The marking results are then pushed to the early warning and intervention module and the stroke prevention and control management platform for intervention and reminders. Potential high-risk individuals include those with risk level 4, risk level 5, and those with at least two abnormal risk characteristics.

Images